LeCun team's counterfactual world models push JEPA from "what happens next" to "what if"

TheTuringPost · x · 2026-11-21

TheTuringPost traces four years of JEPA world models and introduces a new paper by Yann LeCun, Randall Balestriero and Kerda Varaku on Counterfactual World Models.

Timeline of the JEPA lineage:

The new paper's blind spot: a standard JEPA can only answer "what happens if I do this?"—not "what would have happened if I'd done something else?". With a boat pushed by a hidden current, the observed outcome tells you what the current did, but a standard JEPA asked a counterfactual returns the average again. The paper proves a model trained only on factual transitions cannot answer counterfactuals regardless of data or capacity; counterfactual world models instead estimate, in simulation, what an alternative action would have done under the same hidden disturbance.

Related event: LeCun and Colleagues Release Counterfactual World Models Paper(12 posts)→

Original post →

More from Embodied

Embodied channel →